Claims: A processor implemented method for estimating beat to beat interval from noisy Photoplethysmogram (PPG), the method comprising:
receiving a continuous PPG signal and a continuous accelerometer signal captured by one or more wearable sensor worn by a subject (202), wherein the continuous PPG signal is a multichannel PPG signal and the continuous accelerometer signal comprising a x-axis acceleration, a y-axis acceleration and a z-axis acceleration components, wherein the continuous PPG signal comprising Systolic Peaks (SPs) and Dicrotic Peaks (DPs) is corrupted with noise signal, arising primarily from motion;
pre-processing (204):
the multichannel PPG signal to obtain a single channel PPG signal, wherein the single channel PPG signal comprises sequence of a plurality of mean PPG segments; and
the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components of the continuous accelerometer signal to obtain sequence of a plurality of acceleration segments for each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration, and wherein the plurality of mean PPG segments, and the plurality of acceleration segments corresponding to each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components are time synchronized with each other;
modelling noise associated with motion artefacts of the subject with an adaptive Recursive-Least-Square (RLS) filter using the acceleration segments corresponding to each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components as reference signals (206);
eliminating the motion artefacts from each mean PPG segment among the plurality of mean PPG segments to obtain a plurality of partially denoised PPG segments by sequentially subtracting the modelled noise for each acceleration segment of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components (208);
decomposing each partially denoised PPG segment among the plurality of partially denoised segments into a set of Reconstructed Components (RCs) based on Singular Spectrum Analysis (SSA) to obtain most significant oscillatory component-pair corresponding to a highest eigen-value pair (210);
reconstructing a plurality of clean PPG segments using the most significant oscillatory component-pair, wherein the reconstructed plurality of clean PPG segments comprise only the SPs, with the DPs eliminated (212);
applying peak detection to identify the SPs present in each clean PPG segment among the plurality of clean PPG segments (214);
applying peak correction to further refine locations of the SPs identified in each clean PPG segment among the plurality of clean PPG segments (216);
estimating SPSP intervals by determining time interval between every adjacent SP pair of each clean PPG segment (218); and
obtaining valid SPSP intervals of each clean PPG segment, for estimating beat to beat intervals by merging shorter SPSP intervals and discarding incorrigible SPSP intervals from each clean PPG segment using an interval criteria based on a median SPSP interval m_sp (220), wherein the valid SPSP intervals of each clean PPG segment provide the beat to beat interval for seamless monitoring of the subject.

The method as claimed in claim 1, wherein processing the continuous PPG signal comprising the multichannel PPG signal to obtain the single channel PPG signal, comprises:
for each channel of the multichannel PPG signal:
segmenting the continuous PPG signal into a plurality of PPG segments with each PPG segment of a predefined time interval, wherein the predefined time interval is selected to enable real time processing of the plurality of segments for real time estimation of the beat to beat interval;
filtering each PPG segment using a Band Pass Filter (BPF) to eliminate frequency components lying outside a frequency range of interest, wherein the frequency range of interest is determined based on a resting heart rate and a maximum heart rate during extensive exercise of a healthy subject;
normalizing each filtered PPG segment to zero mean and unit variance using a normalization technique to bring the plurality of PPG segments of each channel to the same scale of comparison; and
generating the single channel PPG signal comprising the plurality of mean PPG segments by averaging corresponding PPG segments of each channel that are time synchronized with each other.

The method as claimed in claim 1, wherein preprocessing of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components of the continuous accelerometer signal to obtain sequence of the plurality of acceleration segments for each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration comprises:
segmenting each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components into the plurality of acceleration segments with time interval similar to a predefined interval selected for preprocessing the continuous PPG signal;
filtering each acceleration segment using a Band Pass Filter (BPF) to eliminate frequency components lying outside a frequency range of interest that is similar to a frequency range of interest selected for the preprocessing of the continuous PPG signal; and
normalizing each filtered acceleration segment to zero mean and unit variance using a normalization technique to bring the plurality of acceleration segments of each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components to the same scale of comparison.

The method as claimed in claim 1, wherein the peak correction technique is one of Weighted Local Interpolation, Polynomial Fitting and Global Interpolation.

The method as claimed in claim 1, wherein the interval criteria based on the median SPSP interval m_sp , to obtain valid SPSP intervals for estimating beat to beat intervals comprises:
merging SPSP intervals, of each clean PPG segment, lying below m_sp-0.4*m_sp with one of two adjacent SPSP intervals, which has lower SPSP interval; and
discarding SPSP intervals, of each clean PPG segment, lying above m_sp+0.4*m_sp along with immediately preceding peaks.

A system (100) for estimating beat to beat interval from noisy photoplethysmogram (PPG), comprising:
a memory (102) storing instructions;
one or more Input/Output (I/O) interfaces (106); and
a processor(s) (104) coupled to the memory (102) via the one or more I/O interfaces (106), wherein the processor(s) (104) is configured by the instructions to:
receive a continuous PPG signal and a continuous accelerometer signal captured by one or more wearable sensor worn by a subject, wherein the continuous PPG signal is a multichannel PPG signal and the continuous accelerometer signal comprising a x-axis acceleration, a y-axis acceleration and a z-axis acceleration components, wherein the continuous PPG signal comprising Systolic Peaks (SPs) and Dicrotic Peaks (DPs) is corrupted with noise signal, arising primarily from motion;
pre-process:
the multichannel PPG signal to obtain a single channel PPG signal, wherein the single channel PPG signal comprises sequence of a plurality of mean PPG segments ; and
the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components of the continuous accelerometer signal to obtain sequence of a plurality of acceleration segments for each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration, and wherein the plurality of mean PPG segments, and wherein the plurality of acceleration segments corresponding to each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components are time synchronized with each other;
model noise associated with motion artefacts of the subject with an adaptive Recursive-Least-Square (RLS) filter using the acceleration segments corresponding to each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components as reference signals;
eliminate the motion artefacts from each mean PPG segment among the plurality of mean PPG segments to obtain a plurality of partially denoised PPG segments by sequentially subtracting the modelled noise for each acceleration segment of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components;
decompose each partially denoised PPG segment among the plurality of partially denoised segments into a set of Reconstructed Components (RCs) based on Singular Spectrum Analysis (SSA) to obtain most significant oscillatory component-pair corresponding to a highest eigen-value pair;
reconstruct a plurality of clean PPG segments using the most significant oscillatory component-pair, wherein the reconstructed plurality of clean PPG segments comprise only the SPs, with the DPs eliminated;
apply peak detection to identify the SPs present in each clean PPG segment among the plurality of clean PPG segments;
apply peak correction to further refine locations of the SPs identified in each clean PPG segment among the plurality of clean PPG segments;
estimating SPSP intervals by determining time interval between every adjacent SP pair of each clean PPG segment; and
obtain valid SPSP intervals of each clean PPG segment, for estimating beat to beat intervals by merging shorter SPSP intervals and discarding incorrigible SPSP intervals from each clean PPG segment using an interval criteria based on a median SPSP interval m_sp, wherein the valid SPSP intervals of each clean PPG segment provide the beat to beat interval for seamless monitoring of the subject.

The system (100) as claimed in claim 6, wherein the processor(s) (104) is configured to process the continuous PPG signal comprising the multichannel PPG signal to obtain the single channel PPG signal by performing:
for each channel of the multichannel PPG signal:
segmenting the continuous PPG signal into a plurality of PPG segments with each PPG segment of a predefined time interval, wherein the predefined time interval is selected to enable real time processing of the plurality of segments for real time estimation of the beat to beat interval;
filtering each PPG segment using a Band Pass Filter (BPF) to eliminate frequency components lying outside a frequency range of interest, wherein the frequency range of interest is determined based on a resting heart rate and a maximum heart rate during extensive exercise of a healthy subject;
normalizing each filtered PPG segment to zero mean and unit variance using a normalization technique to bring the plurality of PPG segments of each channel to the same scale of comparison; and
generating the single channel PPG signal comprising the plurality of mean PPG segments by averaging corresponding PPG segments of each channel that are time synchronized with each other.

The system (100) as claimed in claim 6, wherein the processor(s) (104) is configured to preprocess the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components of the continuous accelerometer signal by:
segmenting each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components into the plurality of acceleration segments with time interval similar to a predefined interval selected for preprocessing the continuous PPG signal ;
filtering each acceleration segment using a Band Pass Filter (BPF) to eliminate frequency components lying outside a frequency range of interest that is similar to a frequency range of interest selected for the preprocessing of the continuous PPG signal; and
normalizing each filtered acceleration segment to zero mean and unit variance using a normalization technique to bring the plurality of acceleration segments of each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components to the same scale of comparison.
The system (100) as claimed in claim 6, wherein the processor(s) 104 is configured to utilize the peak correction technique from one of Weighted Local Interpolation, Polynomial Fitting and Global Interpolation.

The system (100) as claimed in claim 6, wherein the interval criteria based on the median SPSP interval m_sp , to obtain valid SPSP intervals for estimating beat to beat intervals comprises:
merging SPSP intervals, of each clean PPG segment, lying below m_sp-0.4*m_sp with one of two adjacent SPSP intervals, which has lower SPSP interval; and
discarding SPSP intervals, of each clean PPG segment, lying above m_sp+0.4*m_sp along with immediately preceding peaks.
, Description:FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003

COMPLETE SPECIFICATION
(See Section 10 and Rule 13)

Title of invention:
METHOD AND SYSTEM FOR ESTIMATING BEAT TO BEAT INTERVALS FROM NOISY PPG

Applicant:
Tata Consultancy Services Limited
A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th Floor,
Nariman Point, Mumbai 400021,
Maharashtra, India

The following specification particularly describes the invention and the manner in which it is to be performed.
TECHNICAL FIELD
The disclosure herein generally relates to beat to beat interval estimation, and, more particularly to beat to beat interval estimation from noisy Photoplethysmogram (PPG) using time domain analysis.

BACKGROUND
Seamless health monitoring with non-invasive approaches is an essential area of research. Wearable sensors are popular choice for continuous health monitoring solutions. Specifically, in cardiac health monitoring, the wearable sensors are used to sense different types of cardiac signals such as Phonocardiogram (PCG), Electrocardiogram (ECG) and Photoplethysmogram (PPG). These signals are used to detect and monitor diseases such as Coronary Artery Disease (CAD) and the like. However, for detection of diseases related to cardiac rhythm such as Atrial Fibrillation (AF), the most important requirement is precise calculation of beat-to-beat cardiac cycles. Thus, ECG signal is an obvious choice for beat-to-beat interval estimation. Though, single-channel ECGs are available and are user friendly to some extent, in this approach the subject needs to set a personal alarm to touch the electrodes in the wearable/ handheld devices equipped with sensors in order to capture the subject’s ECG. Hence, a continuous ECG may prove to be an overkill for the hand-held device, also restricting the person’s movements heavily. Comparatively easier to obtain is a PPG signal which is essentially a volumetric measure of human body fluid, also indicating blood volume through wrist, with physiologically the systolic peaks of the PPG signal following the ECG signal R-peaks. This enables use of the PPG for estimation of beat-to-beat interval. However, primary concern in PPG-based beat-to-beat interval estimation is that the PPG signal from wearable sensors is highly prone to noise due to motion artefacts, in addition to other noises such as white noise and the like. Regular day-to-day activities can heavily distort the signal, thereby making precise systolic peak detection very challenging.

SUMMARY
Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. For example, in one embodiment, there is provided a method for estimating beat to beat interval from noisy Photoplethysmogram (PPG). The method comprises receiving a continuous PPG signal and a continuous accelerometer signal captured by one or more wearable sensor worn by a subject. The continuous PPG signal is a multichannel PPG signal and the continuous accelerometer signal comprising an x-axis acceleration, a y-axis acceleration and a z-axis acceleration components, wherein the continuous PPG signal comprising Systolic Peaks (SPs) and Dicrotic Peaks (DPs) is corrupted with noise signal, arising primarily from motion. The method further comprises pre-processing the multichannel PPG signal to obtain a single channel PPG signal, wherein the single channel PPG signal comprises sequence of a plurality of mean PPG segments and the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components of the continuous accelerometer signal to obtain sequence of a plurality of acceleration segments for each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration. The plurality of mean PPG segments, and the plurality of acceleration segments corresponding to each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components are time synchronized with each other. Further, the method comprises modelling noise associated with motion artefacts of the subject with an adaptive Recursive-Least-Square (RLS) filter using the acceleration segments corresponding to each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components as reference signals. Further, the method comprises eliminating the motion artefacts from each mean PPG segment among the plurality of mean PPG segments to obtain a plurality of partially denoised PPG segments by sequentially subtracting the modelled noise for each acceleration segment of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components. Further, the method comprises decomposing each partially denoised PPG segment among the plurality of partially denoised segments into a set of Reconstructed Components (RCs) based on Singular Spectrum Analysis (SSA) to obtain most significant oscillatory component-pair corresponding to a highest eigen-value pair. Further, the method comprises reconstructing a plurality of clean PPG segments using the most significant oscillatory component-pair, wherein the reconstructed plurality of clean PPG segments comprise only the SPs, with the DPs eliminated. Further, the method comprises applying peak detection to identify the SPs present in each clean PPG segment among the plurality of clean PPG segments. Further, the method comprises applying peak correction to further refine locations of the SPs identified in each clean PPG segment among the plurality of clean PPG segments. Furthermore, the method comprises estimating SPSP intervals by determining time interval between every adjacent SP pair of each clean PPG segment. Furthermore, the method comprises obtaining valid SPSP intervals of each clean PPG segment, for estimating beat to beat intervals by handling shorter SPSP intervals and discarding incorrigible SPSP intervals from each clean PPG segment using an interval criteria based on a median SPSP intervalm_sp, wherein the valid SPSP intervals of each clean PPG segment provide the beat to beat interval for seamless monitoring of the subject.
In yet another aspect, there is provided a system for estimating beat to beat interval from noisy photoplethysmogram (PPG). The system comprising a memory storing instructions; one or more Input/output (I/O) interfaces; and processor(s) coupled to the memory via the one or more I/O interfaces, wherein the processor(s) is configured by the instructions to receive a continuous PPG signal and a continuous accelerometer signal captured by one or more wearable sensor worn by a subject. The continuous PPG signal is a multichannel PPG signal and the continuous accelerometer signal comprising a x-axis acceleration, a y-axis acceleration and a z-axis acceleration components, wherein the continuous PPG signal comprising Systolic Peaks (SPs) and Dicrotic Peaks (DPs) is corrupted with noise signal. Further, the processor (s) is configured to pre-process the multichannel PPG signal to obtain a single channel PPG signal, wherein the single channel PPG signal comprises sequence of a plurality of mean PPG segments and the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components of the continuous accelerometer signal to obtain sequence of a plurality of acceleration segments for each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration. The plurality of mean PPG segments, and the plurality of acceleration segments corresponding to each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components are time synchronized with each other. Further, the processor (s) is configured to model noise associated with motion artefacts of the subject with an adaptive Recursive-Least-Square (RLS) filter using the acceleration segments corresponding to each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components as reference signals. Further, the processor (s) is configured to eliminate the motion artefacts from each mean PPG segment among the plurality of mean PPG segments to obtain a plurality of partially denoised PPG segments by sequentially subtracting the modelled noise for each acceleration segment of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components. Further, the processor (s) is configured to decompose each partially denoised PPG segment among the plurality of partially denoised segments into a set of Reconstructed Components (RCs) based on Singular Spectrum Analysis (SSA) to obtain most significant oscillatory component-pair corresponding to a highest eigen-value pair. Further, the processor (s) is configured to reconstruct a plurality of clean PPG segments using the most significant oscillatory component-pair, wherein the reconstructed plurality of clean PPG segments comprise only the SPs, with the DPs eliminated. Further, the processor (s) is configured to apply peak detection to identify the SPs present in each clean PPG segment among the plurality of clean PPG segments. Further, the processor (s) is configured to apply peak correction to further refine locations of the SPs identified in each clean PPG segment among the plurality of clean PPG segments. Furthermore, the processor (s) is configured to estimating SPSP intervals by determining time interval between every adjacent SP pair of each clean PPG segment; and obtain valid SPSP intervals of each clean PPG segment, for estimating beat to beat intervals by handling shorter SPSP intervals and discarding incorrigible SPSP intervals from each clean PPG segment using an interval criteria based on a median SPSP interval m_sp, wherein the valid SPSP intervals of each clean PPG segment provide the beat to beat interval for seamless monitoring of the subject.
In yet another aspect, there are provided one or more non-transitory machine readable information storage mediums comprising one or more instructions, which when executed by one or more hardware processors causes a method for estimating beat to beat interval from noisy photoplethysmogram (PPG). The method comprises receiving a continuous PPG signal and a continuous accelerometer signal captured by one or more wearable sensor worn by a subject. The continuous PPG signal is a multichannel PPG signal and the continuous accelerometer signal comprising an x-axis acceleration, a y-axis acceleration and a z-axis acceleration components, wherein the continuous PPG signal comprising Systolic Peaks (SPs) and Dicrotic Peaks (DPs) is corrupted with noise signal, arising primarily from motion. The method further comprises pre-processing the multichannel PPG signal to obtain a single channel PPG signal, wherein the single channel PPG signal comprises sequence of a plurality of mean PPG segments and the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components of the continuous accelerometer signal to obtain sequence of a plurality of acceleration segments for each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration. The plurality of mean PPG segments, and the plurality of acceleration segments corresponding to each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components are time synchronized with each other. Further, the method comprises modelling noise associated with motion artefacts of the subject with an adaptive Recursive-Least-Square (RLS) filter using the acceleration segments corresponding to each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components as reference signals. Further, the method comprises eliminating the motion artefacts from each mean PPG segment among the plurality of mean PPG segments to obtain a plurality of partially denoised PPG segments by sequentially subtracting the modelled noise for each acceleration segment of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components. Further, the method comprises decomposing each partially denoised PPG segment among the plurality of partially denoised segments into a set of Reconstructed Components (RCs) based on Singular Spectrum Analysis (SSA) to obtain most significant oscillatory component-pair corresponding to a highest eigen-value pair. Further, the method comprises reconstructing a plurality of clean PPG segments using the most significant oscillatory component-pair, wherein the reconstructed plurality of clean PPG segments comprise only the SPs, with the DPs eliminated. Further, the method comprises applying peak detection to identify the SPs present in each clean PPG segment among the plurality of clean PPG segments. Further, the method comprises applying peak correction to further refine locations of the SPs identified in each clean PPG segment among the plurality of clean PPG segments. Furthermore, the method comprises estimating SPSP intervals by determining time interval between every adjacent SP pair of each clean PPG segment. Furthermore, the method comprises obtaining valid SPSP intervals of each clean PPG segment, for estimating beat to beat intervals by handling shorter SPSP intervals and discarding incorrigible SPSP intervals from each clean PPG segment using an interval criteria based on a median SPSP intervalm_sp, wherein the valid SPSP intervals of each clean PPG segment provide the beat to beat interval for seamless monitoring of the subject.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.

BRIEF DESCRIPTION OF THE DRAWINGS
The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles:
FIG. 1 is a functional block diagram of a system for estimating beat to beat interval from noisy Photoplethysmogram (PPG) signal using time domain analysis, in accordance with some embodiments of the present disclosure.
FIG. 2A and FIG. 2B is a flow diagram illustrating a method for estimating beat to beat interval from the noisy PPG signal by applying time domain analysis using the system of FIG. 1, in accordance with some embodiments of the present disclosure.
FIG. 3 illustrates an example of time-synchronous signals including a gold standard ECG signal, noisy multichannel PPG signals and multiple components of an acceleration signal, in accordance with some embodiments of the present disclosure.
FIG. 4 illustrates effects of various stages of the system of FIG. 1 on the noisy multichannel PPG signals while processing the noisy multichannel PPG signal for estimating beat-to-beat intervals, in accordance with some embodiments of the present disclosure.
FIG. 5 illustrates a block diagram of an adaptive Recursive-Least-Square (RLS) filter used by the system of FIG. 1 for motion artefact removal, in accordance with some embodiments of the present disclosure.
FIG. 6 illustrates most significant oscillatory component-pairs obtained by applying Single Spectrum Analysis (SSA) on PPG signal output of the adaptive RLS filtering stage of the system of FIG. 1, in accordance with some embodiments of the present disclosure.
FIG. 7 illustrates effect of the outlier removal on a reconstructed clean PPG signal, wherein reconstruction is performed using the most significant oscillatory component pair, in accordance with some embodiments of the present disclosure.
FIG. 8A through 8C illustrates Scatter plots and Bland-Altman plots between ECG RR and PPG SPSP intervals for different motion segments, in accordance with some embodiments of the present disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS
Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the scope of the disclosed embodiments. It is intended that the following detailed description be considered as exemplary only, with the true scope being indicated by the following claims.
Several Photoplethysmogram (PPG) motion artefact removal works are present in the literature which exploit acceleration signals. Zhang et al. proposed the TROIKA framework, where a noisy PPG signal was decomposed by Singular Spectrum Analysis (SSA). A clean PPG was reconstructed after rejecting the motion artefact components by frequency domain comparison between PPG components obtained by SSA and simultaneous acceleration signals. Later, Zhang proposed an improved framework JOSS, which jointly estimated the sparse spectra of noisy PPG and acceleration signals. Spectrum subtraction was then performed to eliminate the motion components from PPG spectrum. In another prior work the authors combined adaptive Recursive-Least-Square (RLS) filtering with SSA to develop a hybrid PPG motion artefact removal algorithm. Cascaded Normalized Least Mean Square adaptive filtering was used by Xie et al. Xiong et al. on the other hand, modeled motion artefact removal as a Multi-Channel Spectral Matrix Decomposition optimization problem. Though all of the above-cited prior literature dealt with PPG motion artefact removal, they did not attempt to identify beat-to-beat/ SPSP intervals. Rather their aim was to obtain average heart rate in a time window, for which they mostly adapted frequency domain approaches. Sun et al. performed multi-scale PPG signal analysis using Empirical Mode Decomposition and Hilbert Transform for motion artefact reduction. The systolic peaks (SPs) were then located by finding the positive-to-negative sign changes of the derivative of the cleaned PPG. Han and Kim employed adaptive Least Mean Square (LMS) filtering for PPG motion artefact cancellation and a downward zero crossing based peak detection algorithm for systolic peak identification. Jang et al. filtered the noisy PPG by cascaded recursive digital filters and utilized the Slope Sum Function for peak detection. Moreover, they handled the wrongly identified and missed peaks by rule-based post processing. Vadrevu et al. decomposed the motion contaminated PPG by Stationary Wavelet Transform and computed Multi-scale Sums and Products (MSP) of selected PPG sub-bands to enhance the systolic peaks and suppress the noise. Peak detection on Shannon entropy envelope of MSP was then performed by Gaussian derivative filtering and positive zero-crossing detection.
True signal components and motion artefacts present in PPG are easily separable in the frequency domain. Most of the prior works utilize this observation. However, frequency domain analysis can only be done as a whole over a certain time-window of the signal. In that case frequency domain approach loses the information of individual beats, which is mandatory information for detecting AF. However, processing the PPG in time domain for beat to beat interval estimation is challenging as time domain signal morphology gets heavily distorted by motion artefacts. Thus, detecting individual beats or beat-to-beat intervals is highly difficult using time domain analysis and hardly any successful attempt is observed for precise beat-to-beat/ SPSP interval detection and estimation from noisy PPG providing continuous or real time measurements for seamless cardiac monitoring.
The embodiments herein provide a method and system for estimating beat to beat interval from noisy Photoplethysmogram (PPG) by determining the Systolic Peaks (SP) to SP interval for identifying valid SPSP intervals. The noisy PPG herein refers to a clean PPG signal distorted or corrupted with unwanted signals, primarily due to motion artefacts along with other noise signals. The method disclosed provides a pipeline of signal preprocessing, motion artefact removal, peak detection, peak correction and outlier removal for SPSP interval extraction from the noisy PPG. The method utilizes various post-processing techniques for peak correction. To determine the valid SPSP intervals, the method utilizes an outlier detection approach that merges shorter SPSP intervals and discards incorrigible SPSP intervals from the PPG.
Unlike most existing methods that provide approaches to process PPG signals for cardiac health monitoring with aim to estimate the average heart rate over some pre-defined windows, the method disclosed provides precise detection and estimation of individual beat-to-beat intervals by performing peak detection and subsequent peak correction of the pre-processed PPG and denoised PPG signal in time domain. .
Referring now to the drawings, and more particularly to FIGS. 1 through 8, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary system and/or method.
FIG. 1 is a functional block diagram of a system for estimating beat to beat interval from noisy Photoplethysmogram (PPG) signal using time domain analysis, in accordance with some embodiments of the present disclosure.
In an embodiment, the system 100 includes a processor(s) 104, a communication interface device(s), alternatively referred as input/output (I/O) interface(s) 106, and one or more data storage devices or memory 102 operatively coupled to the processor(s) 104. The processors(s) 104, can be one or more hardware processors. In an embodiment, the one or more hardware processors can be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the processor(s) is configured to fetch and execute computer-readable instructions stored in the memory. In an embodiment, the system 100 can be implemented in a variety of computing systems, such as laptop computers, notebooks, hand-held devices, workstations, mainframe computers, servers, a network cloud and the like.
The I/O interface(s) 106 can include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like and can facilitate multiple communications within a wide variety of networks N/W and protocol types, including wired networks, for example, LAN, cable, etc., and wireless networks, such as WLAN, cellular, or satellite. In an embodiment, the I/O interface device(s) can include one or more ports for connecting a number of devices to one another or to another server. The I/O interface 106 provides interface to connect with one or more wearable sensors worn by a subject, such as S1 through Sn, through which the system 100 acquires a continuous PPG signal and a continuous accelerometer signal of the subject. The continuous PPG signal is a multichannel PPG signal and the continuous accelerometer signal comprises an x-axis acceleration, a y-axis acceleration and a z-axis acceleration component. The memory 102 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes. In an embodiment the memory 102 includes models (not shown) such as the adaptive RLS filter, further explained in conjunction with FIG. 5. The memory 102, stores the continuous PPG signal, the continuous accelerometer signal received from wearable sensors S1 through Sn, preprocessed and processed PPG and accelerometer signals. Thus, the memory 102 may comprise information pertaining to input(s)/output(s) of each step performed by the processor(s) 104 of the system 100 and methods of the present disclosure.
FIG. 2A and FIG. 2B is a flow diagram illustrating a method 200 for estimating the beat to beat interval from the noisy PPG signal by applying time domain analysis using the system of FIG. 1, in accordance with some embodiments of the present disclosure.
In an embodiment, the system 100 comprises one or more data storage devices or the memory 102 operatively coupled to the processor(s) 104 and is configured to store instructions for execution of steps of the method 200 by the processor (s) 104. The steps of the method 200 of the present disclosure will now be explained with reference to the components or blocks of the system 100 as depicted in FIG. 1 and the steps of flow diagram as depicted in FIG. 2. Although process steps, method steps, techniques or the like may be described in a sequential order, such processes, methods and techniques may be configured to work in alternate orders. In other words, any sequence or order of steps that may be described does not necessarily indicate a requirement that the steps to be performed in that order. The steps of processes described herein may be performed in any order practical. Further, some steps may be performed simultaneously.
At step 202 of the method 200, the processor(s) 104 receives a continuous PPG signal and a continuous accelerometer signal captured by one or more wearable sensor worn by a subject to be monitored. As depicted in FIG. 1, the continuous PPG signal is the multichannel PPG signal and the continuous accelerometer signal comprises the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components received from the wearable sensors. The received continuous PPG signal comprises the standard Systolic Peaks (SPs) and Dicrotic Peaks (DPs) but is also corrupted with noise signal, arising primarily from motion. The noise in the PPG from wearable sensors is majorly introduced due to motion or movement of the subject while performing his/her activities of daily living along with other noise sources adding further noise.
FIG. 3 depicts example set of signals obtained from publicly available training dataset of the IEEE Signal Processing Cup 2015. This dataset contains synchronous two-channel wrist-PPG (NP1, NP2) (b) and (c), three-axis accelerometer (Ax, Ay, Az) (d) , (e) and (f) and single-channel chest-ECG (Eg) data (a) of 12 subjects recorded at 125 Hz sampling frequency. During data capture, subjects ran on a treadmill while the speed was varied as follows: rest (30 s) ? 6-8 km/hr (1 min) ? 12-15 km/hr (1 min) ? 6-8 km/hr (1 min) ? 12-15 km/hr (1 min) ? rest (30 s). The subjects moved the hand (equipped with PPG and accelerometer sensors) with the wrist-band purposely, even during rest, to introduce brief motion artefacts. The high activities in the acceleration channels confirm that the primary noise present in the captured PPG arises from motion. The ECG being captured by wet sensor that sticks firmly to the subject’s chest, this signal is much less susceptible to motion artefacts than PPG. Hence ECG can be treated as the gold standard for further comparisons and analysis.
Referring to the steps of the method 200, at step 204, the processor (s) 104 pre-processes the multichannel PPG signal and the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components of the continuous accelerometer signal. Referring to FIG. 3, multichannel PPG signals NP1(b) and NP2 (c) along with the three-axis accelerometer signals Ax (d), Ay (e) and Az(f) are pre-processed.
The sub step of step 204 that pre-processes the multichannel PPG signal to obtain a single channel PPG comprising sequence of a plurality of mean PPG segments is described below. For each channel of the multichannel PPG signal, the pre-processing includes:
Segmenting the continuous PPG signal into a plurality of PPG segments with each PPG segment of a predefined time interval. The predefined time interval is selected to enable real time processing of the plurality of segments for real time estimation of the beat to beat interval. For example, the predefined time interval of 30 sec divides the continuous PPG signal of each channel into consecutive segments of 30 sec. The 30 sec interval enables the method disclosed to be applicable for real-time application.
Filtering each PPG segment using a Band Pass Filter (BPF) to eliminate frequency components lying outside a frequency range of interest, wherein the frequency range of interest is determined based on a resting heart rate and a maximum heart rate during extensive exercise of a healthy subject; The resting heart rate of a healthy adult can vary in 40-100 bpm (approximately 0.6 to 1.6 Hz). The heart rate can rise up to around 200 bpm (approximately 3.3 Hz) during extensive physical activities. Hence, the PPG segments are filtered in 0.6-3.3 Hz by a 10th order Infinite Impulse Response (IIR) Butterworth Band-Pass Filter (BPF) to remove components outside this frequency band of interest. The filter is applied forward and backward to avoid any introduction of phase shift.
Normalizing each filtered PPG segment to zero mean and unit variance using a normalization technique to bring the plurality of PPG segments of each channel to the same scale of comparison. In an embodiment, the filtered PPG signals are normalized to zero mean and unit variance by Z-score normalization to bring all signals to the same scale of comparison. Based on experimental results, the Z-score normalization works well for this method and enables removal of a dc component of the signal and normalizes the variance.
Generating the single channel PPG signal comprising the plurality of mean PPG segments obtained by averaging corresponding PPG segments of each channel that are time synchronized with each other. Since multiple PPG channels are available, their mean is considered for the subsequent steps. In an embodiment, if only single channel PPG is available, the method does not require the averaging step. However, multichannel PPG enables reducing the common noise during pre-processing step.
The preprocessed form Ppre of the noisy PPG of FIG. 3, NP1 (b) and NP2 (c), is shown in FIG. 4(b). It can be observed that several spurious peaks are present in the PPG waveform. Moreover, comparison with the RR intervals obtained from Eg as seen in FIG. 4 (a) suggests that estimates of SPSP intervals are far from accurate. Therefore only conventional preprocessing is not sufficient for removing the artefacts of concern.
Referring to the steps of the method 200, at step 204, the processor (s) 104 also pre-processes the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components of the continuous accelerometer signal to obtain sequence of a plurality of acceleration segments for each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration. The sub steps for preprocessing accelerometer signals include:
Segmenting each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components into the plurality of acceleration segments with time interval similar to the predefined interval selected for preprocessing the continuous PPG signal.
Filtering each acceleration segment using a Band Pass Filter (BPF) to eliminate frequency components lying outside a frequency range of interest that is similar to a frequency range of interest selected for the preprocessing of the continuous PPG signal.
Normalizing each filtered acceleration segment to zero mean and unit variance using a normalization technique to bring the plurality of acceleration segments of each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components to the same scale of comparison;
The plurality of mean PPG segments, and the plurality of acceleration segments corresponding to each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components are time synchronized with each other.
Referring to the steps of the method 200, at step 206, the processor (s) 104 models noise associated with motion artefacts of the subject with an adaptive Recursive-Least-Square (RLS) filter using the acceleration segments corresponding to each of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components as reference signals. The modelled noise is then used for elimination motion artefacts.
Referring to the steps of the method 200, at step 208, the processor (s) 104 eliminates the motion artefacts from each mean PPG segment among the plurality of mean PPG segments to obtain a plurality of partially denoised PPG segments by sequentially subtracting the modelled noise for each acceleration segment of the x-axis acceleration, the y-axis acceleration and the z-axis acceleration components.
Steps 206 and step 208 are explained in conjunction with FIG. 5, wherein FIG. 5 depicts a block diagram of the adaptive Recursive-Least-Square (RLS) filter used by the system of FIG. 1 for motion artefact removal. Motion artefacts are known to be additive to the PPG signal. Adaptive filtering algorithms, e.g. LMS, RLS, perform well in such additive noise cancellation applications. Here, an adaptive Finite Impulse Response (FIR) filter employing conventional RLS technique is used as the first step of removing motion interferences from the PPG signal. RLS is chosen over LMS due to its higher convergence speed and FIR filter is used because of the inherent stability. Firstly, RLS filtering is performed on the preprocessed PPG signal to suppress the artefacts resulting from motion along x-axis by using the x-axis acceleration as the reference signal. Subsequently, the output residual PPG signal is processed twice more by this adaptive RLS filtering using the y-axis and z-axis accelerations sequentially as the references. The filter order and forgetting factor used by the method 200 are empirically chosen as 3 and 0.999 respectively. FIG. 4 (c) shows the waveform Prls obtained after subjecting the preprocessed PPG Ppre of FIG. 4 (b) to adaptive RLS filtering. It is to be noted that, though the SPSP interval estimates are still inaccurate, a significant fraction of the spurious peaks are attenuated by a remarkable margin.
Referring to the steps of the method 200, at step 210, the processor (s) 104 decomposes each partially denoised PPG segment among the plurality of partially denoised segments into a set of Reconstructed Components (RCs) based on Singular Spectrum Analysis (SSA) to obtain most significant oscillatory component-pair corresponding to a highest eigen-value pair (210). In order to have profound motion artefact removal performance, the PPG signal Prls obtained after adaptive RLS filtering is decomposed into the set of Reconstructed Components (RCs) by SSA. Here a window length of 2 sec is used for constructing the trajectory matrix. SSA decomposes a signal into trend, oscillation and noise components. The oscillatory components appear in pairs, both of which represent the same frequency. Corresponding to each pair of oscillatory components, there exists a pair of eigen-values of the covariance of trajectory matrix, which have nearly equal magnitudes. Higher the relative magnitude of the eigen-value pair, more significant is the corresponding oscillation in the signal. Unlike existing approaches, the method disclosed utilizes only the most significant oscillatory component pair obtained by SSA to reconstruct a clean PPG. Thus, the approach disclosed in the method reduces additional processing that may have been required in case all components were considered, still the method disclosed achieves accuracy using the highest component pair.
FIG. 6 illustrates the RCs corresponding to the six highest eigen-values of Prls. The highest eigen-value pair is approximately 2 times the next. Therefore, RC1-2 are much more significant oscillatory components than the rest. Hence, in the method disclosed the system 100 considers the RCs corresponding to the two highest eigen-values. Since the preprocessed PPG signal undergoes adaptive noise cancellation before SSA decomposition, the dominant motion artefacts are already minimized to a remarkable extent, as can be seen in FIG. 4(c). Thus, the most significant component remaining in the PPG signal is the cardiac rhythm itself, which the method captures by identifying the most significant oscillatory component-pair obtained from SSA.
Referring to the steps of the method 200, at step 212, the processor (s) 104 reconstructs a plurality of clean PPG segments using the most significant oscillatory component-pair. Each clean PPG segment among the plurality of clean PPG segments is sinusoidal in nature, wherein the dicrotic peaks (DPs) are mostly eliminated. With the DPs eliminated, the method reduces possibility of error introduction in the systolic peak (SP) detection procedure, by reducing possibility of DPs picked up as SPs. The clean PPG signal P_ssaobtained after SSA processing, along with the detected peaks are shown in FIG. 4 (d). Visibly the SPSP intervals obtained at this stage relate closely to the ground truth RR intervals of FIG. 4 (a).
Referring to the steps of the method 200, at step 214, the processor (s) 104 applies peak detection to identify the SPs present in each clean PPG segment among the plurality of clean PPG segments. Peak detection is performed by identifying locations on clean PPG segments (clean PPG signal) having amplitude larger than those of the neighboring samples on either side, wherein the detected peaks are identified as the systolic peaks (SPs) of the signal. The differences between the locations of consecutive peaks reflects the SPSP intervals.
Once SP locations are identified, referring to the steps of the method 200, at step 216, the processor(s) 104 applies peak correction to further refine locations of the SPs identified or detected in each clean PPG segment among the plurality of clean PPG segments. An example peak corrected PPG signal Pcorr is shown in FIG 4 (e). The precision of the systolic peak locations detected by peak detection only is limited by the sampling frequency (fs) of the captured signal, which is 125 Hz in an example case here. According to the Nyquist Theorem, fs should be at least twice the maximum frequency present in the signal in order to faithfully capture all the details. Hence, fs (= 125 Hz) results into a faithful error resolution of 2 x 1/ fs (= 16 ms), which can be significantly high for low fs. This method disclosed applies one of the following methods for peak correction.
Weighted Local Interpolation: In this method, the location li of the ith systolic peak of the cleaned PPG Pssa is refined by considering its 5 nearest neighboring data points on either side, as given below.
l_i=(?_(k=-5)^5¦?P_ssa (l_i+k)* (l_i+k) ?)/(?_(k=-5)^5¦?P_ssa (l_i+k) ?) (1)
Polynomial Fitting: In this peak correction approach as well the same segment of P_ssa is extracted which contains the concerned systolic peak (SP) location l_i and its 5 nearest neighboring data-points on either side. The coefficients of the best fit 3rd order polynomial for this extracted segment are computed. The real roots of the derivative of this polynomial are determined and the one corresponding to maximum amplitude of P_ssa is chosen as the refined systolic peak.
Global Interpolation: Prior to performing peak detection, the clean PPG segment P_ssais interpolated to 1000 Hz by cubic spline interpolation. Increasing the sampling frequency beyond 1000 Hz seems unnecessary as the performance enhancement is not significant.
All the peak detection methods achieve very similar performances. However, the computational burden of weighted local interpolation is the least. It requires approximately 3.95x 109 instructions to process a 30s data segment whereas polynomial fitting and global interpolation take approximately 4.72x109 and approximately 4.39x109 instructions respectively. The refined SPSP intervals obtained from the illustrative PPG of FIG. 4 (d) by weighted local interpolation are shown in FIG. 4 (e).
Once the peaks are rightly identified, referring to the steps of the method 200, at step 218, the processor (s) 104 estimates SPSP intervals by determining time interval between every adjacent SP pair of each clean PPG segment. However, output of this step may include few SPSP intervals that may be corrupted or distorted and are not good enough to be used for precise estimation of SPSP interval. Referring to the steps of the method 200, at step 204, the processor (s) 104 obtains valid SPSP intervals of each clean PPG segment, for estimating beat to beat intervals by merging shorter SPSP intervals and discarding incorrigible SPSP intervals from each clean PPG segment using an interval criteria based on a median SPSP intervalm_sp, wherein the valid SPSP intervals of each clean PPG segment provide the beat to beat interval for seamless monitoring of the subject. The interval criteria comprises:
1. Merging SPSP intervals, of each clean PPG segment, lying below m_sp-0.4*m_sp with one of two adjacent SPSP intervals, which has lower SPSP interval.
2. Discarding SPSP intervals, of each clean PPG segment, lying above m_sp+0.4*m_sp along with immediately preceding peaks.
Obtaining the valid SPSP intervals refers to applying outlier removal techniques to recognize the incorrigible SPSP intervals. This steps enables to get rid of SP cycles which are beyond correction and if not removed, tend to introduce considerable error in beat-to-beat interval estimation. In an example outlier removal process, the computed SPSP intervals are subjected to the process of outlier removal over 6 sec windows with 2 sec overlap. Let the median SPSP interval of any window be m_sp. The intervals lying below m_sp-0.4*m_sp are considered to be probable errors and are merged with that adjacent SP interval which has lower duration among the two. On the other hand, intervals lying above m_sp+0.4*m_sp along with the immediately preceding peaks are discarded. The corresponding portions of the PPG signal or PPG segments are marked as incorrigible.
As suggested by prior studies, the beat-to-beat intervals can vary by approximately 18% for healthy individuals. The method disclosed imposes a much more relaxed constraint of 40% in order to allow the cycles arising from arrhythmia while discarding only motion-affected cycles which are beyond repair. In the example signal graphs of FIG. 7, the dotted PPG segment is tagged as incorrigible and not considered for beat to beat interval estimation. It can be seen that the corresponding SPSP interval is found as 960.21 ms. This is significantly different from the rest intervals (607.85 - 671.81 ms), which is unlikely for a healthy subject. Hence the proposed algorithm can correctly identify the PPG regions which could not be reconstructed.
EXPERIMENTAL RESULTS:
A. Validation Protocol: ECG R-peaks identified by the modified PanTompkin’s algorithm are used as the ground truth for experimental validation. The improper R-peak identifications are manually corrected wherever possible. The characteristic points of PPG lag those of ECG. Hence, each PPG systolic peak (SP) should be positioned somewhere in between two consecutive R-peaks. Based on this fact, the following rules are developed for performance evaluation of the system 100 and the method 200 disclosed.
1) If no PPG peak is identified between a pair of R-peaks, then a False Negative (FN) occurs.
2) If a single PPG peak is detected between a pair of R-peaks, then it is regarded as a correctly identified peak or a True Positive (TP).
3) If multiple PPG peaks are identified between a pair of R-peaks, then the first peak is regarded as a TP and all other peaks are regarded as False Positives (FP).
Choosing a performance metric is non-trivial in this problem. Hence, literature study is followed to choose Detection Error Rate (DER), Precision (PR) and Recall (RC), which are defined below.
DER = (FP+FN)/TS (2)
PR = TP/(TP+FP) (3)
RC = TP/(TP+FN) (4)
Where, TS is the total number of R-peaks present in ECG. The mean (?DER?_m, ?PR?_m,?RC?_m) and standard deviation (?DER?_s, ?PR?_s,?RC?_s) of the metrics over all subjects are mentioned. For a correctly identified (TP) peak, the SPSP interval following the peak is compared with the immediately preceding ground truth RR interval and the mean and standard deviation of the absolute error between the two, denoted by ( ?AE?_m) and ( ?AE?_s) respectively, are evaluated. Further the average number of instructions (?INS?_m) required to process each subject’s data is obtained using the perfstat command in Linux and the same is reported as a measure of the computational complexity of the method disclosed. In order to avoid the burden of comparing multiple performance metrics for selecting the best method, a Composite Metric (CM) is provided as follows.
CM=?_i¦m_i/max???(M?_i)? (5)
Where, M_(1= ) ?DER?_m, M_(2= ) ?DER?_s ,M_(3= ) ?AE?_m ,M_(4= ) ?AE?_s and M_(5= ) ?INS?_m
Evidently lower the value of CM, better is the performance of the method. Here metrics PR and RC are not considered in the definition of CM given in (5), as minimization of DER ensures their maximization.
B. Performance Comparison: The comparison is performed between four approaches or stages of the method disclosed. The stages used for comparison are referred as:
a) Baseline1 (comprising preprocessing and peak detection)
b) Baseline 2 (comprising Baseline 1 and Motion Artefact Removal using RLS and SSA )
c) Baseline 3 (Baseline 2 and Peak correction)
d) Final method 200 ( Baseline 3 and outlier removal).
Table I below suggests that CM gradually improves from Baseline 1 to the final method. In Baseline 2 better result is obtained when RLS and SSA are used in unison than any one of the two in individual. Moreover different peak correction methods of Baseline 3 are found to produce marginal improvements over Baseline 2. Among these, weighted local interpolation yields the best metric values while incurring the least computational complexity. Though introduction of the outlier removal block in the Final Method increases ?DER?_m and ?DER?_s by 1.03 and 1.07 times respectively, it reduces ?AE?_m and ?AE?_s by 1.04 and 1.08 times respectively. This effect is attributed to the identification and subsequent rejection of the incorrigible parts of the PPG signal. The reductions in ?AE?_m and ?AE?_s outweigh the increases in ?DER?_m and ?DER?_s and CM is found to be the least (92x 105) in case of this method. Thus the last row of Table I presents the final method, which seems to be the best among the methods of table 1, for PPG-based beat-to-beat interval extraction.
Table 1:
Approaches/ stages ?DER?_m ( ?DER?_s)
% ?PR?_m ?(PR?_s)
% ?RC?_m ?(RC?_s)
% ?AE?_m ?(AE?_s)ms ?INS?_m
?×10?^10 CM
?×10?^(-5)
Baseline 1 14.68 (5.82) 89.13 (4.20) 97.46 (3.79) 70.50 (96.91) 2.26 45k
Baseline 2 Only RLS 11.25 (4.62) 91.26 (2.98) 98.26 (2.90) 57.30 (98.87) 2.98 30k
Only SSA 7.90 (7.86) 94.76 (4.99) 97.84 (6.39) 37.34 (70.32) 2.71 15k
RLS+SSA 1.72 (1.48) 98.91 (0.90) 99.38 (0.65) 11.89 (31.59) 3.41 112
Baseline 3 Weighted Local Interpolation 1.63 (1.33) 98.96 (0.84) 99.43 (0.57) 11.80 (31.10) 3.42 94
Polynomial Fitting 1.66 (1.25) 98.91 (0.79) 99.43 (0.54) 11.89 (31.50) 3.61 97
Interpolation 1.71 (1.20) 98.83 (0.77) 99.47 (0.52) 12.43 (35.66) 3.49 110
Final method 1.68 (1.42) 99.05 (0.86) 99.27 (0.60) 11.32 (28.78) 3.45 92

Thus, further result analysis is performed by consideration of the final method 200.
C. Segment-wise Performance Analysis
The performance metrics during different motion segments are shown in table 2 below.
Table 2:
Segment ?DER?_m ( ?DER?_s)
% ?PR?_m ?(PR?_s)
% ?RC?_m ?(RC?_s)
% ?AE?_m ?(AE?_s)ms
Rest 0.98
(1.13)
99.54
(0.65)
99.49
(0.88)
12.37
(23.58)
6-8 KM/Hr -Run 2.10
(1.99)
98.62
(1.33)
99.31
(0.72)
13.84
(35.18)
12-15 KM/Hr -Run 1.25
(1.73)
99.39
(0.92)
99.36
(0.82)
7.85
(15.63)

The significantly high values of PR and RC for all segments confirm that the algorithm is unbiased. It can be observed that the obtained ?AE?_m and ?AE?_s values are the best during fast run, followed by rest and slow run respectively. High speed running involves rhythmic movements of the body. Hence it introduces rhythmic interference in the PPG which is easy to remove by motion artefact removal procedures. Thus, the superior performance metrics for this segment are justified. Though the rest segment should ideally produce the best performance, lower values of ?AE?_m and ?AE?_s are attained for this segment. This is because of the fact that the rest dataset contains brief hand movements which are random, thereby diminishing the performance of the motion artefact removal module. Lastly, the relatively inferior performance for slow run segment is attributed to the randomness of body movements during the slow running event.
FIG. 8A (a and b), FIG 8B (b and c) and FIG. 8C ( e and f) show scatter plots and Bland-Altman (BA) plots between the ECG RR intervals and the PPG SPSP intervals during different motion segments. FIG. 8A (a), FIG. 8B (c), FIG. 8C (e) depict scatter plots and FIG. 8A (b), FIG. 8B (d), FIG. 8C (f) depict BA plots between RR and SPSP intervals for different motion segments. As depicted in FIG. 8A through 8C, s = Slope of linear fit, r = Pearson correlation, p = P-value, b = bias and LOA = Line of agreement. Though most of the mentioned plot parameters suggest that the proposed algorithm can track SPSP intervals from noisy PPG, the significantly high ranges of Line of agreement (LOA) of BA plots indicate
Majority of the existing works on noisy PPG had the goal of average heart-rate estimation over a certain time period (e.g. 8 sec window with 6 sec overlap) for which primarily frequency domain analysis were performed. Those works did not attempt to identify the beat-to-beat cardiac cycles in time-domain, which is approach of the method disclosed. Table 3 below depicts comparison of results of method 200 with existing methods utilizing time-domain approaches particularly using adaptive LMS filtering using acceleration signals for cardiac monitoring. The table 3 reflects that the proposed approach significantly outperforms the state-of-the-art techniques.
Table 3:
Approaches Signals used ?DER?_m ( ?DER?_s)
% ?PR?_m ?(PR?_s)
% ?RC?_m ?(RC?_s)
% ?AE?_m ?(AE?_s)ms
Han et al.
PPG and Acc
27.46
(6.28)
85.84
(3.26)
86.53
(3.57)
64.97
(86.28)
Jang et al. PPG 21.14
(3.99)
83.84
(3.26)
97.97
(3.13)
94.36
(111.86)
Vadrevu et al. PPG 39.65
(4.80) 77.55
(4.35) 85.14
(2.63) 164.83
(135.13)
Method disclosed PPG and Acc 1.68
(1.42)
99.05
(0.86)
99.27
(0.60)
11.32
(28.78)

The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims.
It is to be understood that the scope of the protection is extended to such a program and in addition to a computer-readable means having a message therein; such computer-readable storage means contain program-code means for implementation of one or more steps of the method, when the program runs on a server or mobile device or any suitable programmable device. The hardware device can be any kind of device which can be programmed including e.g. any kind of computer like a server or a personal computer, or the like, or any combination thereof. The device may also include means which could be e.g. hardware means like e.g. an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of hardware and software means, e.g. an ASIC and an FPGA, or at least one microprocessor and at least one memory with software processing components located therein. Thus, the means can include both hardware means and software means. The method embodiments described herein could be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented on different hardware devices, e.g. using a plurality of CPUs.
The embodiments herein can comprise hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc. The functions performed by various components described herein may be implemented in other components or combinations of other components. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
The illustrated steps are set out to explain the exemplary embodiments shown, and it should be anticipated that ongoing technological development will change the manner in which particular functions are performed. These examples are presented herein for purposes of illustration, and not limitation. Further, the boundaries of the functional building blocks have been arbitrarily defined herein for the convenience of the description. Alternative boundaries can be defined so long as the specified functions and relationships thereof are appropriately performed. Alternatives (including equivalents, extensions, variations, deviations, etc., of those described herein) will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein. Such alternatives fall within the scope of the disclosed embodiments. Also, the words “comprising,” “having,” “containing,” and “including,” and other similar forms are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items. It must also be noted that as used herein and in the appended claims, the singular forms “a,” “an,” and “the” include plural references unless the context clearly dictates otherwise.
Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term “computer-readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e., be non-transitory. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.
It is intended that the disclosure and examples be considered as exemplary only, with a true scope of disclosed embodiments being indicated by the following claims.